[gmx-users] How many times should I do Umbrella Sampling

Tsjerk Wassenaar tsjerkw at gmail.com
Thu Jun 9 06:24:12 CEST 2016


Hi Agnivo,

In addition to the remarks of Andre, if you can show that another US run on
the same system gives pretty much the same result, then you can present
that as proof of the robustness, and it is futile to do three more,
especially if you have convergence and sufficient overlap. You can
highlight the regions where the difference is largest, and can probably
argue why a difference of X in those regions will not invalidate your
conclusions.

Another possibility is to construct the PMF using the first 50%, 60%, 70%,
80%, 90%, and 100% of the simulations. Then you can show how the profile
converges and how longer simulations wouldn't help (provided they
wouldn't). That is sort of the same as extending, and I think it's better
than repeating.

Cheers,

Tsjerk
On Jun 9, 2016 4:15 AM, "André Farias de Moura" <moura at ufscar.br> wrote:

> Hi Agnivo,
>
> good overlap of histograms and a smooth pmf cannot be taken as evidences of
> convergence. Some people would extend the simulations to prove that pmf
> does not change appreciably with further sampling, while other people (as
> your reviewer) would do extra independent pmf profiles to prove that all
> profiles agree within a certain precision (and then the best profile would
> be the average, preferably the Jarzynski average of those profiles). I
> prefer the later, though I think that it is a matter of taste, as long as
> you are able to demonstrate that a thorough sampling has been reached. How
> many extra replicas you should do is something you should validate for your
> system (5 replica is somewhat arbitrary).
>
> Maybe you might want to take a look at our papers dealing with that:
>
> J. Phys. Chem. B, 2013, 117 (24), pp 7324–7334
> Phys. Chem. Chem. Phys., 2015,17, 3820-3831
>
> I hope it helps.
>
> Andre
>
>
> On Wed, Jun 8, 2016 at 8:38 PM, Agnivo Gosai <agnivogromacs14 at gmail.com>
> wrote:
>
> > Hello Users
> >
> > I did one umbrella sampling on a trajectory generated by a pulling
> > simulation. There were 45 windows in my US and the PMF curve converged
> with
> > good overlap of histograms.
> >
> > However the reviewer is asking for 5 US simulations from 5 pulling sims
> on
> > the same system and wants the average free energy from those 5 sims. I
> was
> > assuming that the bootstrap calculation takes care of the averages.
> >
> > If there is a proper explanation / reply for the reviewer please share.
> >
> > I am unsure as it takes a long time to do one US sim. I did another one
> for
> > the same system but for a different pulling simulation and the result is
> > almost identical. I think as my pulling code is same I am getting more or
> > less same result.
> >
> >
> > Thanks & Regards
> > Agnivo Gosai
> > Grad Student, Iowa State University.
> > --
> > Gromacs Users mailing list
> >
> > * Please search the archive at
> > http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
> > posting!
> >
> > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
> >
> > * For (un)subscribe requests visit
> > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
> > send a mail to gmx-users-request at gromacs.org.
> >
>
>
>
> --
> _____________
>
> Prof. Dr. André Farias de Moura
> Department of Chemistry
> Federal University of São Carlos
> São Carlos - Brazil
> phone: +55-16-3351-8090
> --
> Gromacs Users mailing list
>
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
> posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
> send a mail to gmx-users-request at gromacs.org.


More information about the gromacs.org_gmx-users mailing list